Meta's Muse Wonderful model scored 79% on the Deepsuite 1.1 benchmark, nearly five points ahead of GPT-6 Astra and Opus 5.5